Buckets:
| import"../chunks/DsnmJJEf.js";import{i as Je,h as Be,C as je,H as d,D as t,E as ze,s as Ue,a as p}from"../chunks/D6eQvJIN.js";import{p as Fe,o as We,s as e,f as c,a as i,b as Ze,c as s,d as g,r as a,n as J}from"../chunks/D9TCMYxZ.js";import{T as h}from"../chunks/eoUhp3Ag.js";import{E as m}from"../chunks/D8azpo8f.js";const Ie='{"title":"ConvBERT","local":"convbert","sections":[{"title":"Overview","local":"overview","sections":[],"depth":2},{"title":"Usage tips","local":"usage-tips","sections":[],"depth":2},{"title":"Resources","local":"resources","sections":[],"depth":2},{"title":"ConvBertConfig","local":"transformers.ConvBertConfig","sections":[],"depth":2},{"title":"ConvBertTokenizer","local":"transformers.ConvBertTokenizer","sections":[],"depth":2},{"title":"ConvBertModel","local":"transformers.ConvBertModel","sections":[],"depth":2},{"title":"ConvBertForMaskedLM","local":"transformers.ConvBertForMaskedLM","sections":[],"depth":2},{"title":"ConvBertForSequenceClassification","local":"transformers.ConvBertForSequenceClassification","sections":[],"depth":2},{"title":"ConvBertForMultipleChoice","local":"transformers.ConvBertForMultipleChoice","sections":[],"depth":2},{"title":"ConvBertForTokenClassification","local":"transformers.ConvBertForTokenClassification","sections":[],"depth":2},{"title":"ConvBertForQuestionAnswering","local":"transformers.ConvBertForQuestionAnswering","sections":[],"depth":2}],"depth":1}';var xe=g('<meta name="hf:doc:metadata"/>'),f=g("<p>Example:</p> <!>",1),u=g(`<p>Although the recipe for forward pass needs to be defined within this function, one should call the <code>Module</code> instance afterwards instead of this since the former takes care of running the pre and post processing steps while | |
| the latter silently ignores them.</p>`),Ge=g("<p>Example of single-label classification:</p> <!>",1),Ne=g("<p>Example of multi-label classification:</p> <!>",1),Re=g(`<p></p> <p><em>This model was published in HF papers on 2020-08-06 and contributed to Hugging Face Transformers on 2021-01-27.</em></p> <!> <!> <!> <p>The ConvBERT model was proposed in <a href="https://huggingface.co/papers/2008.02496" rel="nofollow">ConvBERT: Improving BERT with Span-based Dynamic Convolution</a> by Zihang Jiang, Weihao Yu, Daquan Zhou, Yunpeng Chen, Jiashi Feng, Shuicheng | |
| Yan.</p> <p>The abstract from the paper is the following:</p> <p><em>Pre-trained language models like BERT and its variants have recently achieved impressive performance in various | |
| natural language understanding tasks. However, BERT heavily relies on the global self-attention block and thus suffers | |
| large memory footprint and computation cost. Although all its attention heads query on the whole input sequence for | |
| generating the attention map from a global perspective, we observe some heads only need to learn local dependencies, | |
| which means the existence of computation redundancy. We therefore propose a novel span-based dynamic convolution to | |
| replace these self-attention heads to directly model local dependencies. The novel convolution heads, together with the | |
| rest self-attention heads, form a new mixed attention block that is more efficient at both global and local context | |
| learning. We equip BERT with this mixed attention design and build a ConvBERT model. Experiments have shown that | |
| ConvBERT significantly outperforms BERT and its variants in various downstream tasks, with lower training cost and | |
| fewer model parameters. Remarkably, ConvBERTbase model achieves 86.4 GLUE score, 0.7 higher than ELECTRAbase, while | |
| using less than 1/4 training cost. Code and pre-trained models will be released.</em></p> <p>This model was contributed by <a href="https://huggingface.co/abhishek" rel="nofollow">abhishek</a>. The original implementation can be found | |
| here: <a href="https://github.com/yitu-opensource/ConvBert" rel="nofollow">https://github.com/yitu-opensource/ConvBert</a></p> <!> <p>ConvBERT training tips are similar to those of BERT. For usage tips refer to <a href="bert">BERT documentation</a>.</p> <!> <ul><li><a href="../tasks/sequence_classification">Text classification task guide</a></li> <li><a href="../tasks/token_classification">Token classification task guide</a></li> <li><a href="../tasks/question_answering">Question answering task guide</a></li> <li><a href="../tasks/masked_language_modeling">Masked language modeling task guide</a></li> <li><a href="../tasks/multiple_choice">Multiple choice task guide</a></li></ul> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>This is the configuration class to store the configuration of a ConvBertModel. It is used to instantiate a Convbert | |
| model according to the specified arguments, defining the model architecture. Instantiating a configuration with the | |
| defaults will yield a similar configuration to that of the <a href="https://huggingface.co/YituTech/conv-bert-base" rel="nofollow">YituTech/conv-bert-base</a></p> <p>Configuration objects inherit from <a href="/docs/transformers/pr_49554/en/main_classes/configuration#transformers.PreTrainedConfig">PreTrainedConfig</a> and can be used to control the model outputs. Read the | |
| documentation from <a href="/docs/transformers/pr_49554/en/main_classes/configuration#transformers.PreTrainedConfig">PreTrainedConfig</a> for more information.</p> <!></div> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>Construct a ConvBERT tokenizer (backed by HuggingFace’s tokenizers library). Based on WordPiece.</p> <p>This tokenizer inherits from <a href="/docs/transformers/pr_49554/en/model_doc/layoutlm#transformers.BertTokenizer">BertTokenizer</a> which contains most of the main methods. Users should | |
| refer to this superclass for more information regarding those methods.</p> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>Retrieve sequence ids from a token list that has no special tokens added.</p> <p>For fast tokenizers, data collators call this with <code>already_has_special_tokens=True</code> to build a mask over an | |
| already-formatted sequence. In that case, we compute the mask by checking membership in <code>all_special_ids</code>.</p></div> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!></div></div> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The bare Convbert Model outputting raw hidden-states without any specific head on top.</p> <p>This model inherits from <a href="/docs/transformers/pr_49554/en/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a>. Check the superclass documentation for the generic methods the | |
| library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads | |
| etc.)</p> <p>This model is also a PyTorch <a href="https://pytorch.org/docs/stable/nn.html#torch.nn.Module" rel="nofollow">torch.nn.Module</a> subclass. | |
| Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage | |
| and behavior.</p> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The <a href="/docs/transformers/pr_49554/en/model_doc/convbert#transformers.ConvBertModel">ConvBertModel</a> forward method, overrides the <code>__call__</code> special method.</p> <!> <ul><li><p><strong>last_hidden_state</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length, hidden_size)</code>) — Sequence of hidden-states at the output of the last layer of the model.</p></li> <li><p><strong>hidden_states</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_hidden_states=True</code> is passed or when <code>config.output_hidden_states=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for the output of the embeddings, if the model has an embedding layer, + | |
| one for the output of each layer) of shape <code>(batch_size, sequence_length, hidden_size)</code>.</p> <p>Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.</p></li> <li><p><strong>attentions</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for each layer) of shape <code>(batch_size, num_heads, sequence_length, sequence_length)</code>.</p> <p>Attentions weights after the attention softmax, used to compute the weighted average in the self-attention | |
| heads.</p></li> <li><p><strong>cross_attentions</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_attentions=True</code> and <code>config.add_cross_attention=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for each layer) of shape <code>(batch_size, num_heads, sequence_length, sequence_length)</code>.</p> <p>Attentions weights of the decoder’s cross-attention layer, after the attention softmax, used to compute the | |
| weighted average in the cross-attention heads.</p></li></ul></div></div> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The Convbert Model with a <code>language modeling</code> head on top.”</p> <p>This model inherits from <a href="/docs/transformers/pr_49554/en/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a>. Check the superclass documentation for the generic methods the | |
| library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads | |
| etc.)</p> <p>This model is also a PyTorch <a href="https://pytorch.org/docs/stable/nn.html#torch.nn.Module" rel="nofollow">torch.nn.Module</a> subclass. | |
| Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage | |
| and behavior.</p> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The <a href="/docs/transformers/pr_49554/en/model_doc/convbert#transformers.ConvBertForMaskedLM">ConvBertForMaskedLM</a> forward method, overrides the <code>__call__</code> special method.</p> <!> <ul><li><p><strong>loss</strong> (<code>torch.FloatTensor</code> of shape <code>(1,)</code>, <em>optional</em>, returned when <code>labels</code> is provided) — Masked language modeling (MLM) loss.</p></li> <li><p><strong>logits</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length, config.vocab_size)</code>) — Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).</p></li> <li><p><strong>hidden_states</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_hidden_states=True</code> is passed or when <code>config.output_hidden_states=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for the output of the embeddings, if the model has an embedding layer, + | |
| one for the output of each layer) of shape <code>(batch_size, sequence_length, hidden_size)</code>.</p> <p>Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.</p></li> <li><p><strong>attentions</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for each layer) of shape <code>(batch_size, num_heads, sequence_length, sequence_length)</code>.</p> <p>Attentions weights after the attention softmax, used to compute the weighted average in the self-attention | |
| heads.</p></li></ul> <!></div></div> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>ConvBERT Model transformer with a sequence classification/regression head on top (a linear layer on top of the | |
| pooled output) e.g. for GLUE tasks.</p> <p>This model inherits from <a href="/docs/transformers/pr_49554/en/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a>. Check the superclass documentation for the generic methods the | |
| library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads | |
| etc.)</p> <p>This model is also a PyTorch <a href="https://pytorch.org/docs/stable/nn.html#torch.nn.Module" rel="nofollow">torch.nn.Module</a> subclass. | |
| Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage | |
| and behavior.</p> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The <a href="/docs/transformers/pr_49554/en/model_doc/convbert#transformers.ConvBertForSequenceClassification">ConvBertForSequenceClassification</a> forward method, overrides the <code>__call__</code> special method.</p> <!> <ul><li><p><strong>loss</strong> (<code>torch.FloatTensor</code> of shape <code>(1,)</code>, <em>optional</em>, returned when <code>labels</code> is provided) — Classification (or regression if config.num_labels==1) loss.</p></li> <li><p><strong>logits</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, config.num_labels)</code>) — Classification (or regression if config.num_labels==1) scores (before SoftMax).</p></li> <li><p><strong>hidden_states</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_hidden_states=True</code> is passed or when <code>config.output_hidden_states=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for the output of the embeddings, if the model has an embedding layer, + | |
| one for the output of each layer) of shape <code>(batch_size, sequence_length, hidden_size)</code>.</p> <p>Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.</p></li> <li><p><strong>attentions</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for each layer) of shape <code>(batch_size, num_heads, sequence_length, sequence_length)</code>.</p> <p>Attentions weights after the attention softmax, used to compute the weighted average in the self-attention | |
| heads.</p></li></ul> <!> <!></div></div> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The Convbert Model with a multiple choice classification head on top (a linear layer on top of the pooled output and a | |
| softmax) e.g. for RocStories/SWAG tasks.</p> <p>This model inherits from <a href="/docs/transformers/pr_49554/en/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a>. Check the superclass documentation for the generic methods the | |
| library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads | |
| etc.)</p> <p>This model is also a PyTorch <a href="https://pytorch.org/docs/stable/nn.html#torch.nn.Module" rel="nofollow">torch.nn.Module</a> subclass. | |
| Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage | |
| and behavior.</p> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The <a href="/docs/transformers/pr_49554/en/model_doc/convbert#transformers.ConvBertForMultipleChoice">ConvBertForMultipleChoice</a> forward method, overrides the <code>__call__</code> special method.</p> <!> <ul><li><p><strong>loss</strong> (<code>torch.FloatTensor</code> of shape <em>(1,)</em>, <em>optional</em>, returned when <code>labels</code> is provided) — Classification loss.</p></li> <li><p><strong>logits</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, num_choices)</code>) — <em>num_choices</em> is the second dimension of the input tensors. (see <em>input_ids</em> above).</p> <p>Classification scores (before SoftMax).</p></li> <li><p><strong>hidden_states</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_hidden_states=True</code> is passed or when <code>config.output_hidden_states=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for the output of the embeddings, if the model has an embedding layer, + | |
| one for the output of each layer) of shape <code>(batch_size, sequence_length, hidden_size)</code>.</p> <p>Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.</p></li> <li><p><strong>attentions</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for each layer) of shape <code>(batch_size, num_heads, sequence_length, sequence_length)</code>.</p> <p>Attentions weights after the attention softmax, used to compute the weighted average in the self-attention | |
| heads.</p></li></ul> <!></div></div> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The Convbert transformer with a token classification head on top (a linear layer on top of the hidden-states | |
| output) e.g. for Named-Entity-Recognition (NER) tasks.</p> <p>This model inherits from <a href="/docs/transformers/pr_49554/en/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a>. Check the superclass documentation for the generic methods the | |
| library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads | |
| etc.)</p> <p>This model is also a PyTorch <a href="https://pytorch.org/docs/stable/nn.html#torch.nn.Module" rel="nofollow">torch.nn.Module</a> subclass. | |
| Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage | |
| and behavior.</p> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The <a href="/docs/transformers/pr_49554/en/model_doc/convbert#transformers.ConvBertForTokenClassification">ConvBertForTokenClassification</a> forward method, overrides the <code>__call__</code> special method.</p> <!> <ul><li><p><strong>loss</strong> (<code>torch.FloatTensor</code> of shape <code>(1,)</code>, <em>optional</em>, returned when <code>labels</code> is provided) — Classification loss.</p></li> <li><p><strong>logits</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length, config.num_labels)</code>) — Classification scores (before SoftMax).</p></li> <li><p><strong>hidden_states</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_hidden_states=True</code> is passed or when <code>config.output_hidden_states=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for the output of the embeddings, if the model has an embedding layer, + | |
| one for the output of each layer) of shape <code>(batch_size, sequence_length, hidden_size)</code>.</p> <p>Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.</p></li> <li><p><strong>attentions</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for each layer) of shape <code>(batch_size, num_heads, sequence_length, sequence_length)</code>.</p> <p>Attentions weights after the attention softmax, used to compute the weighted average in the self-attention | |
| heads.</p></li></ul> <!></div></div> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The Convbert transformer with a span classification head on top for extractive question-answering tasks like | |
| SQuAD (a linear layer on top of the hidden-states output to compute <code>span start logits</code> and <code>span end logits</code>).</p> <p>This model inherits from <a href="/docs/transformers/pr_49554/en/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a>. Check the superclass documentation for the generic methods the | |
| library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads | |
| etc.)</p> <p>This model is also a PyTorch <a href="https://pytorch.org/docs/stable/nn.html#torch.nn.Module" rel="nofollow">torch.nn.Module</a> subclass. | |
| Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage | |
| and behavior.</p> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The <a href="/docs/transformers/pr_49554/en/model_doc/convbert#transformers.ConvBertForQuestionAnswering">ConvBertForQuestionAnswering</a> forward method, overrides the <code>__call__</code> special method.</p> <!> <ul><li><p><strong>loss</strong> (<code>torch.FloatTensor</code> of shape <code>(1,)</code>, <em>optional</em>, returned when <code>labels</code> is provided) — Total span extraction loss is the sum of a Cross-Entropy for the start and end positions.</p></li> <li><p><strong>start_logits</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length)</code>) — Span-start scores (before SoftMax).</p></li> <li><p><strong>end_logits</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length)</code>) — Span-end scores (before SoftMax).</p></li> <li><p><strong>hidden_states</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_hidden_states=True</code> is passed or when <code>config.output_hidden_states=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for the output of the embeddings, if the model has an embedding layer, + | |
| one for the output of each layer) of shape <code>(batch_size, sequence_length, hidden_size)</code>.</p> <p>Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.</p></li> <li><p><strong>attentions</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for each layer) of shape <code>(batch_size, num_heads, sequence_length, sequence_length)</code>.</p> <p>Attentions weights after the attention softmax, used to compute the weighted average in the self-attention | |
| heads.</p></li></ul> <!></div></div> <!> <p></p>`,1);function Qe(ue,ge){Fe(ge,!1),We(()=>{new URLSearchParams(window.location.search).get("fw")}),Je();var B=Re();Be("75fyv9",n=>{var r=xe();Ue(r,"content",Ie),i(n,r)});var j=e(c(B),4);je(j,{containerStyle:"float: right; margin-left: 10px; display: inline-flex; position: relative; z-index: 10;"});var z=e(j,2);d(z,{title:"ConvBERT",local:"convbert",headingTag:"h1"});var U=e(z,2);d(U,{title:"Overview",local:"overview",headingTag:"h2"});var F=e(U,10);d(F,{title:"Usage tips",local:"usage-tips",headingTag:"h2"});var W=e(F,4);d(W,{title:"Resources",local:"resources",headingTag:"h2"});var Z=e(W,4);d(Z,{title:"ConvBertConfig",local:"transformers.ConvBertConfig",headingTag:"h2"});var _=e(Z,2),I=s(_);t(I,{name:"class transformers.ConvBertConfig",anchor:"transformers.ConvBertConfig",source:"https://github.com/huggingface/transformers/blob/vr_49554/src/transformers/models/convbert/configuration_convbert.py#L24",parameters:[{name:"transformers_version",val:": str | None = None"},{name:"architectures",val:": list[str] | None = None"},{name:"output_hidden_states",val:": bool | None = False"},{name:"return_dict",val:": bool | None = True"},{name:"dtype",val:": str | torch.dtype | None = None"},{name:"chunk_size_feed_forward",val:": int = 0"},{name:"is_encoder_decoder",val:": bool = False"},{name:"id2label",val:": dict[int, str] | dict[str, str] | None = None"},{name:"label2id",val:": dict[str, int] | dict[str, str] | None = None"},{name:"problem_type",val:": Literal['regression', 'single_label_classification', 'multi_label_classification'] | None = None"},{name:"vocab_size",val:": int = 30522"},{name:"hidden_size",val:": int = 768"},{name:"num_hidden_layers",val:": int = 12"},{name:"num_attention_heads",val:": int = 12"},{name:"intermediate_size",val:": int = 3072"},{name:"hidden_act",val:": str = 'gelu'"},{name:"hidden_dropout_prob",val:": float | int = 0.1"},{name:"attention_probs_dropout_prob",val:": float | int = 0.1"},{name:"max_position_embeddings",val:": int = 512"},{name:"type_vocab_size",val:": int = 2"},{name:"initializer_range",val:": float = 0.02"},{name:"layer_norm_eps",val:": float = 1e-12"},{name:"pad_token_id",val:": int | None = 1"},{name:"bos_token_id",val:": int | None = 0"},{name:"eos_token_id",val:": int | list[int] | None = 2"},{name:"embedding_size",val:": int = 768"},{name:"head_ratio",val:": int = 2"},{name:"conv_kernel_size",val:": int = 9"},{name:"num_groups",val:": int = 1"},{name:"classifier_dropout",val:": float | int | None = None"},{name:"is_decoder",val:": bool = False"},{name:"add_cross_attention",val:": bool = False"},{name:"tie_word_embeddings",val:": bool = True"}],parametersDescription:[{anchor:"transformers.ConvBertConfig.vocab_size",description:`<strong>vocab_size</strong> (<code>int</code>, <em>optional</em>, defaults to <code>30522</code>) — | |
| Vocabulary size of the model. Defines the number of different tokens that can be represented by the <code>input_ids</code>.`,name:"vocab_size"},{anchor:"transformers.ConvBertConfig.hidden_size",description:`<strong>hidden_size</strong> (<code>int</code>, <em>optional</em>, defaults to <code>768</code>) — | |
| Dimension of the hidden representations.`,name:"hidden_size"},{anchor:"transformers.ConvBertConfig.num_hidden_layers",description:`<strong>num_hidden_layers</strong> (<code>int</code>, <em>optional</em>, defaults to <code>12</code>) — | |
| Number of hidden layers in the Transformer decoder.`,name:"num_hidden_layers"},{anchor:"transformers.ConvBertConfig.num_attention_heads",description:`<strong>num_attention_heads</strong> (<code>int</code>, <em>optional</em>, defaults to <code>12</code>) — | |
| Number of attention heads for each attention layer in the Transformer decoder.`,name:"num_attention_heads"},{anchor:"transformers.ConvBertConfig.intermediate_size",description:`<strong>intermediate_size</strong> (<code>int</code>, <em>optional</em>, defaults to <code>3072</code>) — | |
| Dimension of the MLP representations.`,name:"intermediate_size"},{anchor:"transformers.ConvBertConfig.hidden_act",description:`<strong>hidden_act</strong> (<code>str</code>, <em>optional</em>, defaults to <code>gelu</code>) — | |
| The non-linear activation function (function or string) in the decoder. For example, <code>"gelu"</code>, | |
| <code>"relu"</code>, <code>"silu"</code>, etc.`,name:"hidden_act"},{anchor:"transformers.ConvBertConfig.hidden_dropout_prob",description:`<strong>hidden_dropout_prob</strong> (<code>Union[float, int]</code>, <em>optional</em>, defaults to <code>0.1</code>) — | |
| The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.`,name:"hidden_dropout_prob"},{anchor:"transformers.ConvBertConfig.attention_probs_dropout_prob",description:`<strong>attention_probs_dropout_prob</strong> (<code>Union[float, int]</code>, <em>optional</em>, defaults to <code>0.1</code>) — | |
| The dropout ratio for the attention probabilities.`,name:"attention_probs_dropout_prob"},{anchor:"transformers.ConvBertConfig.max_position_embeddings",description:`<strong>max_position_embeddings</strong> (<code>int</code>, <em>optional</em>, defaults to <code>512</code>) — | |
| The maximum sequence length that this model might ever be used with.`,name:"max_position_embeddings"},{anchor:"transformers.ConvBertConfig.type_vocab_size",description:`<strong>type_vocab_size</strong> (<code>int</code>, <em>optional</em>, defaults to <code>2</code>) — | |
| The vocabulary size of the <code>token_type_ids</code>.`,name:"type_vocab_size"},{anchor:"transformers.ConvBertConfig.initializer_range",description:`<strong>initializer_range</strong> (<code>float</code>, <em>optional</em>, defaults to <code>0.02</code>) — | |
| The standard deviation of the truncated_normal_initializer for initializing all weight matrices.`,name:"initializer_range"},{anchor:"transformers.ConvBertConfig.layer_norm_eps",description:`<strong>layer_norm_eps</strong> (<code>float</code>, <em>optional</em>, defaults to <code>1e-12</code>) — | |
| The epsilon used by the layer normalization layers.`,name:"layer_norm_eps"},{anchor:"transformers.ConvBertConfig.pad_token_id",description:`<strong>pad_token_id</strong> (<code>int</code>, <em>optional</em>, defaults to <code>1</code>) — | |
| Token id used for padding in the vocabulary.`,name:"pad_token_id"},{anchor:"transformers.ConvBertConfig.bos_token_id",description:`<strong>bos_token_id</strong> (<code>int</code>, <em>optional</em>, defaults to <code>0</code>) — | |
| Token id used for beginning-of-stream in the vocabulary.`,name:"bos_token_id"},{anchor:"transformers.ConvBertConfig.eos_token_id",description:`<strong>eos_token_id</strong> (<code>Union[int, list[int]]</code>, <em>optional</em>, defaults to <code>2</code>) — | |
| Token id used for end-of-stream in the vocabulary.`,name:"eos_token_id"},{anchor:"transformers.ConvBertConfig.embedding_size",description:`<strong>embedding_size</strong> (<code>int</code>, <em>optional</em>, defaults to <code>768</code>) — | |
| Dimensionality of the embeddings and hidden states.`,name:"embedding_size"},{anchor:"transformers.ConvBertConfig.head_ratio",description:`<strong>head_ratio</strong> (<code>int</code>, <em>optional</em>, defaults to 2) — | |
| Ratio gamma to reduce the number of attention heads.`,name:"head_ratio"},{anchor:"transformers.ConvBertConfig.conv_kernel_size",description:`<strong>conv_kernel_size</strong> (<code>int</code>, <em>optional</em>, defaults to <code>9</code>) — | |
| The size of the convolutional kernel.`,name:"conv_kernel_size"},{anchor:"transformers.ConvBertConfig.num_groups",description:`<strong>num_groups</strong> (<code>int</code>, <em>optional</em>, defaults to 1) — | |
| The number of groups for grouped linear layers for ConvBert model`,name:"num_groups"},{anchor:"transformers.ConvBertConfig.classifier_dropout",description:`<strong>classifier_dropout</strong> (<code>Union[float, int]</code>, <em>optional</em>) — | |
| The dropout ratio for classifier.`,name:"classifier_dropout"},{anchor:"transformers.ConvBertConfig.is_decoder",description:`<strong>is_decoder</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>False</code>) — | |
| Whether the model is used as a decoder or not. If <code>False</code>, the model is used as an encoder.`,name:"is_decoder"},{anchor:"transformers.ConvBertConfig.add_cross_attention",description:`<strong>add_cross_attention</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>False</code>) — | |
| Whether cross-attention layers should be added to the model.`,name:"add_cross_attention"},{anchor:"transformers.ConvBertConfig.tie_word_embeddings",description:`<strong>tie_word_embeddings</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>True</code>) — | |
| Whether to tie weight embeddings according to model’s <code>tied_weights_keys</code> mapping.`,name:"tie_word_embeddings"}]});var fe=e(I,6);m(fe,{anchor:"transformers.ConvBertConfig.example",children:(n,r)=>{var o=f(),l=e(c(o),2);p(l,{code:"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",highlighted:`<span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> ConvBertConfig, ConvBertModel | |
| <span class="hljs-meta">>>> </span><span class="hljs-comment"># Initializing a ConvBERT convbert-base-uncased style configuration</span> | |
| <span class="hljs-meta">>>> </span>configuration = ConvBertConfig() | |
| <span class="hljs-meta">>>> </span><span class="hljs-comment"># Initializing a model (with random weights) from the convbert-base-uncased style configuration</span> | |
| <span class="hljs-meta">>>> </span>model = ConvBertModel(configuration) | |
| <span class="hljs-meta">>>> </span><span class="hljs-comment"># Accessing the model configuration</span> | |
| <span class="hljs-meta">>>> </span>configuration = model.config`,lang:"python",wrap:!1}),i(n,o)},$$slots:{default:!0}}),a(_);var x=e(_,2);d(x,{title:"ConvBertTokenizer",local:"transformers.ConvBertTokenizer",headingTag:"h2"});var b=e(x,2),G=s(b);t(G,{name:"class transformers.ConvBertTokenizer",anchor:"transformers.ConvBertTokenizer",source:"https://github.com/huggingface/transformers/blob/vr_49554/src/transformers/models/convbert/tokenization_convbert.py#L19",parameters:[{name:"vocab",val:": str | dict[str, int] | None = None"},{name:"do_lower_case",val:": bool = True"},{name:"unk_token",val:": str = '[UNK]'"},{name:"sep_token",val:": str = '[SEP]'"},{name:"pad_token",val:": str = '[PAD]'"},{name:"cls_token",val:": str = '[CLS]'"},{name:"mask_token",val:": str = '[MASK]'"},{name:"tokenize_chinese_chars",val:": bool = True"},{name:"strip_accents",val:": bool | None = None"},{name:"**kwargs",val:""}]});var v=e(G,6),_e=s(v);t(_e,{name:"get_special_tokens_mask",anchor:"transformers.ConvBertTokenizer.get_special_tokens_mask",source:"https://github.com/huggingface/transformers/blob/vr_49554/src/transformers/tokenization_utils_base.py#L1309",parameters:[{name:"token_ids_0",val:": list[int]"},{name:"token_ids_1",val:": list[int] | None = None"},{name:"already_has_special_tokens",val:": bool = False"}],parametersDescription:[{anchor:"transformers.ConvBertTokenizer.get_special_tokens_mask.token_ids_0",description:"<strong>token_ids_0</strong> — List of IDs for the (possibly already formatted) sequence.",name:"token_ids_0"},{anchor:"transformers.ConvBertTokenizer.get_special_tokens_mask.token_ids_1",description:"<strong>token_ids_1</strong> — Unused when <code>already_has_special_tokens=True</code>. Must be None in that case.",name:"token_ids_1"},{anchor:"transformers.ConvBertTokenizer.get_special_tokens_mask.already_has_special_tokens",description:"<strong>already_has_special_tokens</strong> — Whether the sequence is already formatted with special tokens.",name:"already_has_special_tokens"}],returnDescription:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p>1 for a special token, 0 for a sequence token.</p> | |
| `,returnType:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p>A list of integers in the range [0, 1]</p> | |
| `}),J(4),a(v);var N=e(v,2),be=s(N);t(be,{name:"save_vocabulary",anchor:"transformers.ConvBertTokenizer.save_vocabulary",source:"https://github.com/huggingface/transformers/blob/vr_49554/src/transformers/tokenization_utils_tokenizers.py#L545",parameters:[{name:"save_directory",val:": str"},{name:"filename_prefix",val:": str | None = None"}]}),a(N),a(b);var R=e(b,2);d(R,{title:"ConvBertModel",local:"transformers.ConvBertModel",headingTag:"h2"});var y=e(R,2),V=s(y);t(V,{name:"class transformers.ConvBertModel",anchor:"transformers.ConvBertModel",source:"https://github.com/huggingface/transformers/blob/vr_49554/src/transformers/models/convbert/modeling_convbert.py#L576",parameters:[{name:"config",val:""}],parametersDescription:[{anchor:"transformers.ConvBertModel.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_49554/en/model_doc/convbert#transformers.ConvBertModel">ConvBertModel</a>) — | |
| Model configuration class with all the parameters of the model. Initializing with a config file does not | |
| load the weights associated with the model, only the configuration. Check out the | |
| <a href="/docs/transformers/pr_49554/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}]});var q=e(V,8),X=s(q);t(X,{name:"forward",anchor:"transformers.ConvBertModel.forward",source:"https://github.com/huggingface/transformers/blob/vr_49554/src/transformers/models/convbert/modeling_convbert.py#L595",parameters:[{name:"input_ids",val:": typing.Optional[torch.LongTensor] = None"},{name:"attention_mask",val:": typing.Optional[torch.FloatTensor] = None"},{name:"token_type_ids",val:": typing.Optional[torch.LongTensor] = None"},{name:"position_ids",val:": typing.Optional[torch.LongTensor] = None"},{name:"inputs_embeds",val:": typing.Optional[torch.FloatTensor] = None"},{name:"**kwargs",val:": Unpack"}],parametersDescription:[{anchor:"transformers.ConvBertModel.forward.input_ids",description:`<strong>input_ids</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Indices of input sequence tokens in the vocabulary. Padding will be ignored by default.</p> | |
| <p>Indices can be obtained using <a href="/docs/transformers/pr_49554/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_49554/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and | |
| <a href="/docs/transformers/pr_49554/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.__call__">PreTrainedTokenizer.<strong>call</strong>()</a> for details.</p> | |
| <p><a href="../glossary#input-ids">What are input IDs?</a>`,name:"input_ids"},{anchor:"transformers.ConvBertModel.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Mask to avoid performing attention on padding token indices. Mask values selected in <code>[0, 1]</code>:</p> | |
| <ul> | |
| <li>1 for tokens that are <strong>not masked</strong>,</li> | |
| <li>0 for tokens that are <strong>masked</strong>.</li> | |
| </ul> | |
| <p><a href="../glossary#attention-mask">What are attention masks?</a>`,name:"attention_mask"},{anchor:"transformers.ConvBertModel.forward.token_type_ids",description:`<strong>token_type_ids</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Segment token indices to indicate first and second portions of the inputs. Indices are selected in <code>[0, 1]</code>:</p> | |
| <ul> | |
| <li>0 corresponds to a <em>sentence A</em> token,</li> | |
| <li>1 corresponds to a <em>sentence B</em> token.</li> | |
| </ul> | |
| <p><a href="../glossary#token-type-ids">What are token type IDs?</a>`,name:"token_type_ids"},{anchor:"transformers.ConvBertModel.forward.position_ids",description:`<strong>position_ids</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Indices of positions of each input sequence tokens in the position embeddings. Selected in the range <code>[0, config.n_positions - 1]</code>.</p> | |
| <p><a href="../glossary#position-ids">What are position IDs?</a>`,name:"position_ids"},{anchor:"transformers.ConvBertModel.forward.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length, hidden_size)</code>, <em>optional</em>) — | |
| Optionally, instead of passing <code>input_ids</code> you can choose to directly pass an embedded representation. This | |
| is useful if you want more control over how to convert <code>input_ids</code> indices into associated vectors than the | |
| model’s internal embedding lookup matrix.`,name:"inputs_embeds"}],returnDescription:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p>A <a | |
| href="/docs/transformers/pr_49554/en/main_classes/output#transformers.modeling_outputs.BaseModelOutputWithCrossAttentions" | |
| >BaseModelOutputWithCrossAttentions</a> or a tuple of | |
| <code>torch.FloatTensor</code> (if <code>return_dict=False</code> is passed or when <code>config.return_dict=False</code>) comprising various | |
| elements depending on the configuration (<a | |
| href="/docs/transformers/pr_49554/en/model_doc/convbert#transformers.ConvBertConfig" | |
| >ConvBertConfig</a>) and inputs.</p> | |
| `,returnType:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p><a | |
| href="/docs/transformers/pr_49554/en/main_classes/output#transformers.modeling_outputs.BaseModelOutputWithCrossAttentions" | |
| >BaseModelOutputWithCrossAttentions</a> or <code>tuple(torch.FloatTensor)</code></p> | |
| `});var ve=e(X,4);h(ve,{children:(n,r)=>{var o=u();i(n,o)},$$slots:{default:!0}}),J(2),a(q),a(y);var Y=e(y,2);d(Y,{title:"ConvBertForMaskedLM",local:"transformers.ConvBertForMaskedLM",headingTag:"h2"});var T=e(Y,2),E=s(T);t(E,{name:"class transformers.ConvBertForMaskedLM",anchor:"transformers.ConvBertForMaskedLM",source:"https://github.com/huggingface/transformers/blob/vr_49554/src/transformers/models/convbert/modeling_convbert.py#L671",parameters:[{name:"config",val:""}],parametersDescription:[{anchor:"transformers.ConvBertForMaskedLM.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_49554/en/model_doc/convbert#transformers.ConvBertForMaskedLM">ConvBertForMaskedLM</a>) — | |
| Model configuration class with all the parameters of the model. Initializing with a config file does not | |
| load the weights associated with the model, only the configuration. Check out the | |
| <a href="/docs/transformers/pr_49554/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}]});var Q=e(E,8),S=s(Q);t(S,{name:"forward",anchor:"transformers.ConvBertForMaskedLM.forward",source:"https://github.com/huggingface/transformers/blob/vr_49554/src/transformers/models/convbert/modeling_convbert.py#L690",parameters:[{name:"input_ids",val:": typing.Optional[torch.LongTensor] = None"},{name:"attention_mask",val:": typing.Optional[torch.FloatTensor] = None"},{name:"token_type_ids",val:": typing.Optional[torch.LongTensor] = None"},{name:"position_ids",val:": typing.Optional[torch.LongTensor] = None"},{name:"inputs_embeds",val:": typing.Optional[torch.FloatTensor] = None"},{name:"labels",val:": typing.Optional[torch.LongTensor] = None"},{name:"**kwargs",val:": Unpack"}],parametersDescription:[{anchor:"transformers.ConvBertForMaskedLM.forward.input_ids",description:`<strong>input_ids</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Indices of input sequence tokens in the vocabulary. Padding will be ignored by default.</p> | |
| <p>Indices can be obtained using <a href="/docs/transformers/pr_49554/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_49554/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and | |
| <a href="/docs/transformers/pr_49554/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.__call__">PreTrainedTokenizer.<strong>call</strong>()</a> for details.</p> | |
| <p><a href="../glossary#input-ids">What are input IDs?</a>`,name:"input_ids"},{anchor:"transformers.ConvBertForMaskedLM.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Mask to avoid performing attention on padding token indices. Mask values selected in <code>[0, 1]</code>:</p> | |
| <ul> | |
| <li>1 for tokens that are <strong>not masked</strong>,</li> | |
| <li>0 for tokens that are <strong>masked</strong>.</li> | |
| </ul> | |
| <p><a href="../glossary#attention-mask">What are attention masks?</a>`,name:"attention_mask"},{anchor:"transformers.ConvBertForMaskedLM.forward.token_type_ids",description:`<strong>token_type_ids</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Segment token indices to indicate first and second portions of the inputs. Indices are selected in <code>[0, 1]</code>:</p> | |
| <ul> | |
| <li>0 corresponds to a <em>sentence A</em> token,</li> | |
| <li>1 corresponds to a <em>sentence B</em> token.</li> | |
| </ul> | |
| <p><a href="../glossary#token-type-ids">What are token type IDs?</a>`,name:"token_type_ids"},{anchor:"transformers.ConvBertForMaskedLM.forward.position_ids",description:`<strong>position_ids</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Indices of positions of each input sequence tokens in the position embeddings. Selected in the range <code>[0, config.n_positions - 1]</code>.</p> | |
| <p><a href="../glossary#position-ids">What are position IDs?</a>`,name:"position_ids"},{anchor:"transformers.ConvBertForMaskedLM.forward.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length, hidden_size)</code>, <em>optional</em>) — | |
| Optionally, instead of passing <code>input_ids</code> you can choose to directly pass an embedded representation. This | |
| is useful if you want more control over how to convert <code>input_ids</code> indices into associated vectors than the | |
| model’s internal embedding lookup matrix.`,name:"inputs_embeds"},{anchor:"transformers.ConvBertForMaskedLM.forward.labels",description:`<strong>labels</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Labels for computing the masked language modeling loss. Indices should be in <code>[-100, 0, ..., config.vocab_size]</code> (see <code>input_ids</code> docstring) Tokens with indices set to <code>-100</code> are ignored (masked), the | |
| loss is only computed for the tokens with labels in <code>[0, ..., config.vocab_size]</code>.`,name:"labels"}],returnDescription:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p>A <a | |
| href="/docs/transformers/pr_49554/en/main_classes/output#transformers.modeling_outputs.MaskedLMOutput" | |
| >MaskedLMOutput</a> or a tuple of | |
| <code>torch.FloatTensor</code> (if <code>return_dict=False</code> is passed or when <code>config.return_dict=False</code>) comprising various | |
| elements depending on the configuration (<a | |
| href="/docs/transformers/pr_49554/en/model_doc/convbert#transformers.ConvBertConfig" | |
| >ConvBertConfig</a>) and inputs.</p> | |
| `,returnType:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p><a | |
| href="/docs/transformers/pr_49554/en/main_classes/output#transformers.modeling_outputs.MaskedLMOutput" | |
| >MaskedLMOutput</a> or <code>tuple(torch.FloatTensor)</code></p> | |
| `});var L=e(S,4);h(L,{children:(n,r)=>{var o=u();i(n,o)},$$slots:{default:!0}});var ye=e(L,4);m(ye,{anchor:"transformers.ConvBertForMaskedLM.forward.example",children:(n,r)=>{var o=f(),l=e(c(o),2);p(l,{code:"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",highlighted:`<span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer, ConvBertForMaskedLM | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-meta">>>> </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"YituTech/conv-bert-base"</span>) | |
| <span class="hljs-meta">>>> </span>model = ConvBertForMaskedLM.from_pretrained(<span class="hljs-string">"YituTech/conv-bert-base"</span>) | |
| <span class="hljs-meta">>>> </span>inputs = tokenizer(<span class="hljs-string">"The capital of France is <mask>."</span>, return_tensors=<span class="hljs-string">"pt"</span>) | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">with</span> torch.no_grad(): | |
| <span class="hljs-meta">... </span> logits = model(**inputs).logits | |
| <span class="hljs-meta">>>> </span><span class="hljs-comment"># retrieve index of <mask></span> | |
| <span class="hljs-meta">>>> </span>mask_token_index = (inputs.input_ids == tokenizer.mask_token_id)[<span class="hljs-number">0</span>].nonzero(as_tuple=<span class="hljs-literal">True</span>)[<span class="hljs-number">0</span>] | |
| <span class="hljs-meta">>>> </span>predicted_token_id = logits[<span class="hljs-number">0</span>, mask_token_index].argmax(axis=-<span class="hljs-number">1</span>) | |
| <span class="hljs-meta">>>> </span>tokenizer.decode(predicted_token_id) | |
| ... | |
| <span class="hljs-meta">>>> </span>labels = tokenizer(<span class="hljs-string">"The capital of France is Paris."</span>, return_tensors=<span class="hljs-string">"pt"</span>)[<span class="hljs-string">"input_ids"</span>] | |
| <span class="hljs-meta">>>> </span><span class="hljs-comment"># mask labels of non-<mask> tokens</span> | |
| <span class="hljs-meta">>>> </span>labels = torch.where(inputs.input_ids == tokenizer.mask_token_id, labels, -<span class="hljs-number">100</span>) | |
| <span class="hljs-meta">>>> </span>outputs = model(**inputs, labels=labels) | |
| <span class="hljs-meta">>>> </span><span class="hljs-built_in">round</span>(outputs.loss.item(), <span class="hljs-number">2</span>) | |
| ...`,lang:"python",wrap:!1}),i(n,o)},$$slots:{default:!0}}),a(Q),a(T);var H=e(T,2);d(H,{title:"ConvBertForSequenceClassification",local:"transformers.ConvBertForSequenceClassification",headingTag:"h2"});var M=e(H,2),A=s(M);t(A,{name:"class transformers.ConvBertForSequenceClassification",anchor:"transformers.ConvBertForSequenceClassification",source:"https://github.com/huggingface/transformers/blob/vr_49554/src/transformers/models/convbert/modeling_convbert.py#L759",parameters:[{name:"config",val:""}],parametersDescription:[{anchor:"transformers.ConvBertForSequenceClassification.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_49554/en/model_doc/convbert#transformers.ConvBertForSequenceClassification">ConvBertForSequenceClassification</a>) — | |
| Model configuration class with all the parameters of the model. Initializing with a config file does not | |
| load the weights associated with the model, only the configuration. Check out the | |
| <a href="/docs/transformers/pr_49554/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}]});var P=e(A,8),D=s(P);t(D,{name:"forward",anchor:"transformers.ConvBertForSequenceClassification.forward",source:"https://github.com/huggingface/transformers/blob/vr_49554/src/transformers/models/convbert/modeling_convbert.py#L770",parameters:[{name:"input_ids",val:": typing.Optional[torch.LongTensor] = None"},{name:"attention_mask",val:": typing.Optional[torch.FloatTensor] = None"},{name:"token_type_ids",val:": typing.Optional[torch.LongTensor] = None"},{name:"position_ids",val:": typing.Optional[torch.LongTensor] = None"},{name:"inputs_embeds",val:": typing.Optional[torch.FloatTensor] = None"},{name:"labels",val:": typing.Optional[torch.LongTensor] = None"},{name:"**kwargs",val:": Unpack"}],parametersDescription:[{anchor:"transformers.ConvBertForSequenceClassification.forward.input_ids",description:`<strong>input_ids</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Indices of input sequence tokens in the vocabulary. Padding will be ignored by default.</p> | |
| <p>Indices can be obtained using <a href="/docs/transformers/pr_49554/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_49554/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and | |
| <a href="/docs/transformers/pr_49554/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.__call__">PreTrainedTokenizer.<strong>call</strong>()</a> for details.</p> | |
| <p><a href="../glossary#input-ids">What are input IDs?</a>`,name:"input_ids"},{anchor:"transformers.ConvBertForSequenceClassification.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Mask to avoid performing attention on padding token indices. Mask values selected in <code>[0, 1]</code>:</p> | |
| <ul> | |
| <li>1 for tokens that are <strong>not masked</strong>,</li> | |
| <li>0 for tokens that are <strong>masked</strong>.</li> | |
| </ul> | |
| <p><a href="../glossary#attention-mask">What are attention masks?</a>`,name:"attention_mask"},{anchor:"transformers.ConvBertForSequenceClassification.forward.token_type_ids",description:`<strong>token_type_ids</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Segment token indices to indicate first and second portions of the inputs. Indices are selected in <code>[0, 1]</code>:</p> | |
| <ul> | |
| <li>0 corresponds to a <em>sentence A</em> token,</li> | |
| <li>1 corresponds to a <em>sentence B</em> token.</li> | |
| </ul> | |
| <p><a href="../glossary#token-type-ids">What are token type IDs?</a>`,name:"token_type_ids"},{anchor:"transformers.ConvBertForSequenceClassification.forward.position_ids",description:`<strong>position_ids</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Indices of positions of each input sequence tokens in the position embeddings. Selected in the range <code>[0, config.n_positions - 1]</code>.</p> | |
| <p><a href="../glossary#position-ids">What are position IDs?</a>`,name:"position_ids"},{anchor:"transformers.ConvBertForSequenceClassification.forward.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length, hidden_size)</code>, <em>optional</em>) — | |
| Optionally, instead of passing <code>input_ids</code> you can choose to directly pass an embedded representation. This | |
| is useful if you want more control over how to convert <code>input_ids</code> indices into associated vectors than the | |
| model’s internal embedding lookup matrix.`,name:"inputs_embeds"},{anchor:"transformers.ConvBertForSequenceClassification.forward.labels",description:`<strong>labels</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size,)</code>, <em>optional</em>) — | |
| Labels for computing the sequence classification/regression loss. Indices should be in <code>[0, ..., config.num_labels - 1]</code>. If <code>config.num_labels == 1</code> a regression loss is computed (Mean-Square loss), If | |
| <code>config.num_labels > 1</code> a classification loss is computed (Cross-Entropy).`,name:"labels"}],returnDescription:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p>A <a | |
| href="/docs/transformers/pr_49554/en/main_classes/output#transformers.modeling_outputs.SequenceClassifierOutput" | |
| >SequenceClassifierOutput</a> or a tuple of | |
| <code>torch.FloatTensor</code> (if <code>return_dict=False</code> is passed or when <code>config.return_dict=False</code>) comprising various | |
| elements depending on the configuration (<a | |
| href="/docs/transformers/pr_49554/en/model_doc/convbert#transformers.ConvBertConfig" | |
| >ConvBertConfig</a>) and inputs.</p> | |
| `,returnType:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p><a | |
| href="/docs/transformers/pr_49554/en/main_classes/output#transformers.modeling_outputs.SequenceClassifierOutput" | |
| >SequenceClassifierOutput</a> or <code>tuple(torch.FloatTensor)</code></p> | |
| `});var K=e(D,4);h(K,{children:(n,r)=>{var o=u();i(n,o)},$$slots:{default:!0}});var O=e(K,4);m(O,{anchor:"transformers.ConvBertForSequenceClassification.forward.example",children:(n,r)=>{var o=Ge(),l=e(c(o),2);p(l,{code:"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",highlighted:`<span class="hljs-meta">>>> </span><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer, ConvBertForSequenceClassification | |
| <span class="hljs-meta">>>> </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"YituTech/conv-bert-base"</span>) | |
| <span class="hljs-meta">>>> </span>model = ConvBertForSequenceClassification.from_pretrained(<span class="hljs-string">"YituTech/conv-bert-base"</span>) | |
| <span class="hljs-meta">>>> </span>inputs = tokenizer(<span class="hljs-string">"Hello, my dog is cute"</span>, return_tensors=<span class="hljs-string">"pt"</span>) | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">with</span> torch.no_grad(): | |
| <span class="hljs-meta">... </span> logits = model(**inputs).logits | |
| <span class="hljs-meta">>>> </span>predicted_class_id = logits.argmax().item() | |
| <span class="hljs-meta">>>> </span>model.config.id2label[predicted_class_id] | |
| ... | |
| <span class="hljs-meta">>>> </span><span class="hljs-comment"># To train a model on \`num_labels\` classes, you can pass \`num_labels=num_labels\` to \`.from_pretrained(...)\`</span> | |
| <span class="hljs-meta">>>> </span>num_labels = <span class="hljs-built_in">len</span>(model.config.id2label) | |
| <span class="hljs-meta">>>> </span>model = ConvBertForSequenceClassification.from_pretrained(<span class="hljs-string">"YituTech/conv-bert-base"</span>, num_labels=num_labels) | |
| <span class="hljs-meta">>>> </span>labels = torch.tensor([<span class="hljs-number">1</span>]) | |
| <span class="hljs-meta">>>> </span>loss = model(**inputs, labels=labels).loss | |
| <span class="hljs-meta">>>> </span><span class="hljs-built_in">round</span>(loss.item(), <span class="hljs-number">2</span>) | |
| ...`,lang:"python",wrap:!1}),i(n,o)},$$slots:{default:!0}});var Te=e(O,2);m(Te,{anchor:"transformers.ConvBertForSequenceClassification.forward.example-2",children:(n,r)=>{var o=Ne(),l=e(c(o),2);p(l,{code:"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",highlighted:`<span class="hljs-meta">>>> </span><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer, ConvBertForSequenceClassification | |
| <span class="hljs-meta">>>> </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"YituTech/conv-bert-base"</span>) | |
| <span class="hljs-meta">>>> </span>model = ConvBertForSequenceClassification.from_pretrained(<span class="hljs-string">"YituTech/conv-bert-base"</span>, problem_type=<span class="hljs-string">"multi_label_classification"</span>) | |
| <span class="hljs-meta">>>> </span>inputs = tokenizer(<span class="hljs-string">"Hello, my dog is cute"</span>, return_tensors=<span class="hljs-string">"pt"</span>) | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">with</span> torch.no_grad(): | |
| <span class="hljs-meta">... </span> logits = model(**inputs).logits | |
| <span class="hljs-meta">>>> </span>predicted_class_ids = torch.arange(<span class="hljs-number">0</span>, logits.shape[-<span class="hljs-number">1</span>])[torch.sigmoid(logits).squeeze(dim=<span class="hljs-number">0</span>) > <span class="hljs-number">0.5</span>] | |
| <span class="hljs-meta">>>> </span><span class="hljs-comment"># To train a model on \`num_labels\` classes, you can pass \`num_labels=num_labels\` to \`.from_pretrained(...)\`</span> | |
| <span class="hljs-meta">>>> </span>num_labels = <span class="hljs-built_in">len</span>(model.config.id2label) | |
| <span class="hljs-meta">>>> </span>model = ConvBertForSequenceClassification.from_pretrained( | |
| <span class="hljs-meta">... </span> <span class="hljs-string">"YituTech/conv-bert-base"</span>, num_labels=num_labels, problem_type=<span class="hljs-string">"multi_label_classification"</span> | |
| <span class="hljs-meta">... </span>) | |
| <span class="hljs-meta">>>> </span>labels = torch.<span class="hljs-built_in">sum</span>( | |
| <span class="hljs-meta">... </span> torch.nn.functional.one_hot(predicted_class_ids[<span class="hljs-literal">None</span>, :].clone(), num_classes=num_labels), dim=<span class="hljs-number">1</span> | |
| <span class="hljs-meta">... </span>).to(torch.<span class="hljs-built_in">float</span>) | |
| <span class="hljs-meta">>>> </span>loss = model(**inputs, labels=labels).loss`,lang:"python",wrap:!1}),i(n,o)},$$slots:{default:!0}}),a(P),a(M);var $=e(M,2);d($,{title:"ConvBertForMultipleChoice",local:"transformers.ConvBertForMultipleChoice",headingTag:"h2"});var k=e($,2),ee=s(k);t(ee,{name:"class transformers.ConvBertForMultipleChoice",anchor:"transformers.ConvBertForMultipleChoice",source:"https://github.com/huggingface/transformers/blob/vr_49554/src/transformers/models/convbert/modeling_convbert.py#L826",parameters:[{name:"config",val:""}],parametersDescription:[{anchor:"transformers.ConvBertForMultipleChoice.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_49554/en/model_doc/convbert#transformers.ConvBertForMultipleChoice">ConvBertForMultipleChoice</a>) — | |
| Model configuration class with all the parameters of the model. Initializing with a config file does not | |
| load the weights associated with the model, only the configuration. Check out the | |
| <a href="/docs/transformers/pr_49554/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}]});var oe=e(ee,8),ne=s(oe);t(ne,{name:"forward",anchor:"transformers.ConvBertForMultipleChoice.forward",source:"https://github.com/huggingface/transformers/blob/vr_49554/src/transformers/models/convbert/modeling_convbert.py#L837",parameters:[{name:"input_ids",val:": typing.Optional[torch.LongTensor] = None"},{name:"attention_mask",val:": typing.Optional[torch.FloatTensor] = None"},{name:"token_type_ids",val:": typing.Optional[torch.LongTensor] = None"},{name:"position_ids",val:": typing.Optional[torch.LongTensor] = None"},{name:"inputs_embeds",val:": typing.Optional[torch.FloatTensor] = None"},{name:"labels",val:": typing.Optional[torch.LongTensor] = None"},{name:"**kwargs",val:": Unpack"}],parametersDescription:[{anchor:"transformers.ConvBertForMultipleChoice.forward.input_ids",description:`<strong>input_ids</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, num_choices, sequence_length)</code>) — | |
| Indices of input sequence tokens in the vocabulary.</p> | |
| <p>Indices can be obtained using <a href="/docs/transformers/pr_49554/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_49554/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and | |
| <a href="/docs/transformers/pr_49554/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.__call__">PreTrainedTokenizer.<strong>call</strong>()</a> for details.</p> | |
| <p><a href="../glossary#input-ids">What are input IDs?</a>`,name:"input_ids"},{anchor:"transformers.ConvBertForMultipleChoice.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Mask to avoid performing attention on padding token indices. Mask values selected in <code>[0, 1]</code>:</p> | |
| <ul> | |
| <li>1 for tokens that are <strong>not masked</strong>,</li> | |
| <li>0 for tokens that are <strong>masked</strong>.</li> | |
| </ul> | |
| <p><a href="../glossary#attention-mask">What are attention masks?</a>`,name:"attention_mask"},{anchor:"transformers.ConvBertForMultipleChoice.forward.token_type_ids",description:`<strong>token_type_ids</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, num_choices, sequence_length)</code>, <em>optional</em>) — | |
| Segment token indices to indicate first and second portions of the inputs. Indices are selected in <code>[0, 1]</code>:`,name:"token_type_ids"}],returnDescription:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p>A <a | |
| href="/docs/transformers/pr_49554/en/main_classes/output#transformers.modeling_outputs.MultipleChoiceModelOutput" | |
| >MultipleChoiceModelOutput</a> or a tuple of | |
| <code>torch.FloatTensor</code> (if <code>return_dict=False</code> is passed or when <code>config.return_dict=False</code>) comprising various | |
| elements depending on the configuration (<a | |
| href="/docs/transformers/pr_49554/en/model_doc/convbert#transformers.ConvBertConfig" | |
| >ConvBertConfig</a>) and inputs.</p> | |
| `,returnType:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p><a | |
| href="/docs/transformers/pr_49554/en/main_classes/output#transformers.modeling_outputs.MultipleChoiceModelOutput" | |
| >MultipleChoiceModelOutput</a> or <code>tuple(torch.FloatTensor)</code></p> | |
| `});var te=e(ne,4);h(te,{children:(n,r)=>{var o=u();i(n,o)},$$slots:{default:!0}});var Me=e(te,4);m(Me,{anchor:"transformers.ConvBertForMultipleChoice.forward.example",children:(n,r)=>{var o=f(),l=e(c(o),2);p(l,{code:"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",highlighted:`<span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer, ConvBertForMultipleChoice | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-meta">>>> </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"YituTech/conv-bert-base"</span>) | |
| <span class="hljs-meta">>>> </span>model = ConvBertForMultipleChoice.from_pretrained(<span class="hljs-string">"YituTech/conv-bert-base"</span>) | |
| <span class="hljs-meta">>>> </span>prompt = <span class="hljs-string">"In Italy, pizza served in formal settings, such as at a restaurant, is presented unsliced."</span> | |
| <span class="hljs-meta">>>> </span>choice0 = <span class="hljs-string">"It is eaten with a fork and a knife."</span> | |
| <span class="hljs-meta">>>> </span>choice1 = <span class="hljs-string">"It is eaten while held in the hand."</span> | |
| <span class="hljs-meta">>>> </span>labels = torch.tensor(<span class="hljs-number">0</span>).unsqueeze(<span class="hljs-number">0</span>) <span class="hljs-comment"># choice0 is correct (according to Wikipedia ;)), batch size 1</span> | |
| <span class="hljs-meta">>>> </span>encoding = tokenizer([prompt, prompt], [choice0, choice1], return_tensors=<span class="hljs-string">"pt"</span>, padding=<span class="hljs-literal">True</span>) | |
| <span class="hljs-meta">>>> </span>outputs = model(**{k: v.unsqueeze(<span class="hljs-number">0</span>) <span class="hljs-keyword">for</span> k, v <span class="hljs-keyword">in</span> encoding.items()}, labels=labels) <span class="hljs-comment"># batch size is 1</span> | |
| <span class="hljs-meta">>>> </span><span class="hljs-comment"># the linear classifier still needs to be trained</span> | |
| <span class="hljs-meta">>>> </span>loss = outputs.loss | |
| <span class="hljs-meta">>>> </span>logits = outputs.logits`,lang:"python",wrap:!1}),i(n,o)},$$slots:{default:!0}}),a(oe),a(k);var se=e(k,2);d(se,{title:"ConvBertForTokenClassification",local:"transformers.ConvBertForTokenClassification",headingTag:"h2"});var w=e(se,2),ae=s(w);t(ae,{name:"class transformers.ConvBertForTokenClassification",anchor:"transformers.ConvBertForTokenClassification",source:"https://github.com/huggingface/transformers/blob/vr_49554/src/transformers/models/convbert/modeling_convbert.py#L917",parameters:[{name:"config",val:""}],parametersDescription:[{anchor:"transformers.ConvBertForTokenClassification.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_49554/en/model_doc/convbert#transformers.ConvBertForTokenClassification">ConvBertForTokenClassification</a>) — | |
| Model configuration class with all the parameters of the model. Initializing with a config file does not | |
| load the weights associated with the model, only the configuration. Check out the | |
| <a href="/docs/transformers/pr_49554/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}]});var re=e(ae,8),ie=s(re);t(ie,{name:"forward",anchor:"transformers.ConvBertForTokenClassification.forward",source:"https://github.com/huggingface/transformers/blob/vr_49554/src/transformers/models/convbert/modeling_convbert.py#L932",parameters:[{name:"input_ids",val:": typing.Optional[torch.LongTensor] = None"},{name:"attention_mask",val:": typing.Optional[torch.FloatTensor] = None"},{name:"token_type_ids",val:": typing.Optional[torch.LongTensor] = None"},{name:"position_ids",val:": typing.Optional[torch.LongTensor] = None"},{name:"inputs_embeds",val:": typing.Optional[torch.FloatTensor] = None"},{name:"labels",val:": typing.Optional[torch.LongTensor] = None"},{name:"**kwargs",val:": Unpack"}],parametersDescription:[{anchor:"transformers.ConvBertForTokenClassification.forward.input_ids",description:`<strong>input_ids</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Indices of input sequence tokens in the vocabulary. Padding will be ignored by default.</p> | |
| <p>Indices can be obtained using <a href="/docs/transformers/pr_49554/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_49554/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and | |
| <a href="/docs/transformers/pr_49554/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.__call__">PreTrainedTokenizer.<strong>call</strong>()</a> for details.</p> | |
| <p><a href="../glossary#input-ids">What are input IDs?</a>`,name:"input_ids"},{anchor:"transformers.ConvBertForTokenClassification.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Mask to avoid performing attention on padding token indices. Mask values selected in <code>[0, 1]</code>:</p> | |
| <ul> | |
| <li>1 for tokens that are <strong>not masked</strong>,</li> | |
| <li>0 for tokens that are <strong>masked</strong>.</li> | |
| </ul> | |
| <p><a href="../glossary#attention-mask">What are attention masks?</a>`,name:"attention_mask"},{anchor:"transformers.ConvBertForTokenClassification.forward.token_type_ids",description:`<strong>token_type_ids</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Segment token indices to indicate first and second portions of the inputs. Indices are selected in <code>[0, 1]</code>:</p> | |
| <ul> | |
| <li>0 corresponds to a <em>sentence A</em> token,</li> | |
| <li>1 corresponds to a <em>sentence B</em> token.</li> | |
| </ul> | |
| <p><a href="../glossary#token-type-ids">What are token type IDs?</a>`,name:"token_type_ids"},{anchor:"transformers.ConvBertForTokenClassification.forward.position_ids",description:`<strong>position_ids</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Indices of positions of each input sequence tokens in the position embeddings. Selected in the range <code>[0, config.n_positions - 1]</code>.</p> | |
| <p><a href="../glossary#position-ids">What are position IDs?</a>`,name:"position_ids"},{anchor:"transformers.ConvBertForTokenClassification.forward.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length, hidden_size)</code>, <em>optional</em>) — | |
| Optionally, instead of passing <code>input_ids</code> you can choose to directly pass an embedded representation. This | |
| is useful if you want more control over how to convert <code>input_ids</code> indices into associated vectors than the | |
| model’s internal embedding lookup matrix.`,name:"inputs_embeds"},{anchor:"transformers.ConvBertForTokenClassification.forward.labels",description:`<strong>labels</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Labels for computing the token classification loss. Indices should be in <code>[0, ..., config.num_labels - 1]</code>.`,name:"labels"}],returnDescription:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p>A <a | |
| href="/docs/transformers/pr_49554/en/main_classes/output#transformers.modeling_outputs.TokenClassifierOutput" | |
| >TokenClassifierOutput</a> or a tuple of | |
| <code>torch.FloatTensor</code> (if <code>return_dict=False</code> is passed or when <code>config.return_dict=False</code>) comprising various | |
| elements depending on the configuration (<a | |
| href="/docs/transformers/pr_49554/en/model_doc/convbert#transformers.ConvBertConfig" | |
| >ConvBertConfig</a>) and inputs.</p> | |
| `,returnType:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p><a | |
| href="/docs/transformers/pr_49554/en/main_classes/output#transformers.modeling_outputs.TokenClassifierOutput" | |
| >TokenClassifierOutput</a> or <code>tuple(torch.FloatTensor)</code></p> | |
| `});var le=e(ie,4);h(le,{children:(n,r)=>{var o=u();i(n,o)},$$slots:{default:!0}});var ke=e(le,4);m(ke,{anchor:"transformers.ConvBertForTokenClassification.forward.example",children:(n,r)=>{var o=f(),l=e(c(o),2);p(l,{code:"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",highlighted:`<span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer, ConvBertForTokenClassification | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-meta">>>> </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"YituTech/conv-bert-base"</span>) | |
| <span class="hljs-meta">>>> </span>model = ConvBertForTokenClassification.from_pretrained(<span class="hljs-string">"YituTech/conv-bert-base"</span>) | |
| <span class="hljs-meta">>>> </span>inputs = tokenizer( | |
| <span class="hljs-meta">... </span> <span class="hljs-string">"HuggingFace is a company based in Paris and New York"</span>, add_special_tokens=<span class="hljs-literal">False</span>, return_tensors=<span class="hljs-string">"pt"</span> | |
| <span class="hljs-meta">... </span>) | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">with</span> torch.no_grad(): | |
| <span class="hljs-meta">... </span> logits = model(**inputs).logits | |
| <span class="hljs-meta">>>> </span>predicted_token_class_ids = logits.argmax(-<span class="hljs-number">1</span>) | |
| <span class="hljs-meta">>>> </span><span class="hljs-comment"># Note that tokens are classified rather then input words which means that</span> | |
| <span class="hljs-meta">>>> </span><span class="hljs-comment"># there might be more predicted token classes than words.</span> | |
| <span class="hljs-meta">>>> </span><span class="hljs-comment"># Multiple token classes might account for the same word</span> | |
| <span class="hljs-meta">>>> </span>predicted_tokens_classes = [model.config.id2label[t.item()] <span class="hljs-keyword">for</span> t <span class="hljs-keyword">in</span> predicted_token_class_ids[<span class="hljs-number">0</span>]] | |
| <span class="hljs-meta">>>> </span>predicted_tokens_classes | |
| ... | |
| <span class="hljs-meta">>>> </span>labels = predicted_token_class_ids | |
| <span class="hljs-meta">>>> </span>loss = model(**inputs, labels=labels).loss | |
| <span class="hljs-meta">>>> </span><span class="hljs-built_in">round</span>(loss.item(), <span class="hljs-number">2</span>) | |
| ...`,lang:"python",wrap:!1}),i(n,o)},$$slots:{default:!0}}),a(re),a(w);var de=e(w,2);d(de,{title:"ConvBertForQuestionAnswering",local:"transformers.ConvBertForQuestionAnswering",headingTag:"h2"});var C=e(de,2),ce=s(C);t(ce,{name:"class transformers.ConvBertForQuestionAnswering",anchor:"transformers.ConvBertForQuestionAnswering",source:"https://github.com/huggingface/transformers/blob/vr_49554/src/transformers/models/convbert/modeling_convbert.py#L972",parameters:[{name:"config",val:""}],parametersDescription:[{anchor:"transformers.ConvBertForQuestionAnswering.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_49554/en/model_doc/convbert#transformers.ConvBertForQuestionAnswering">ConvBertForQuestionAnswering</a>) — | |
| Model configuration class with all the parameters of the model. Initializing with a config file does not | |
| load the weights associated with the model, only the configuration. Check out the | |
| <a href="/docs/transformers/pr_49554/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}]});var pe=e(ce,8),me=s(pe);t(me,{name:"forward",anchor:"transformers.ConvBertForQuestionAnswering.forward",source:"https://github.com/huggingface/transformers/blob/vr_49554/src/transformers/models/convbert/modeling_convbert.py#L983",parameters:[{name:"input_ids",val:": typing.Optional[torch.LongTensor] = None"},{name:"attention_mask",val:": typing.Optional[torch.FloatTensor] = None"},{name:"token_type_ids",val:": typing.Optional[torch.LongTensor] = None"},{name:"position_ids",val:": typing.Optional[torch.LongTensor] = None"},{name:"inputs_embeds",val:": typing.Optional[torch.FloatTensor] = None"},{name:"start_positions",val:": typing.Optional[torch.LongTensor] = None"},{name:"end_positions",val:": typing.Optional[torch.LongTensor] = None"},{name:"**kwargs",val:": Unpack"}],parametersDescription:[{anchor:"transformers.ConvBertForQuestionAnswering.forward.input_ids",description:`<strong>input_ids</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Indices of input sequence tokens in the vocabulary. Padding will be ignored by default.</p> | |
| <p>Indices can be obtained using <a href="/docs/transformers/pr_49554/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_49554/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and | |
| <a href="/docs/transformers/pr_49554/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.__call__">PreTrainedTokenizer.<strong>call</strong>()</a> for details.</p> | |
| <p><a href="../glossary#input-ids">What are input IDs?</a>`,name:"input_ids"},{anchor:"transformers.ConvBertForQuestionAnswering.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Mask to avoid performing attention on padding token indices. Mask values selected in <code>[0, 1]</code>:</p> | |
| <ul> | |
| <li>1 for tokens that are <strong>not masked</strong>,</li> | |
| <li>0 for tokens that are <strong>masked</strong>.</li> | |
| </ul> | |
| <p><a href="../glossary#attention-mask">What are attention masks?</a>`,name:"attention_mask"},{anchor:"transformers.ConvBertForQuestionAnswering.forward.token_type_ids",description:`<strong>token_type_ids</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Segment token indices to indicate first and second portions of the inputs. Indices are selected in <code>[0, 1]</code>:</p> | |
| <ul> | |
| <li>0 corresponds to a <em>sentence A</em> token,</li> | |
| <li>1 corresponds to a <em>sentence B</em> token.</li> | |
| </ul> | |
| <p><a href="../glossary#token-type-ids">What are token type IDs?</a>`,name:"token_type_ids"},{anchor:"transformers.ConvBertForQuestionAnswering.forward.position_ids",description:`<strong>position_ids</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Indices of positions of each input sequence tokens in the position embeddings. Selected in the range <code>[0, config.n_positions - 1]</code>.</p> | |
| <p><a href="../glossary#position-ids">What are position IDs?</a>`,name:"position_ids"},{anchor:"transformers.ConvBertForQuestionAnswering.forward.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length, hidden_size)</code>, <em>optional</em>) — | |
| Optionally, instead of passing <code>input_ids</code> you can choose to directly pass an embedded representation. This | |
| is useful if you want more control over how to convert <code>input_ids</code> indices into associated vectors than the | |
| model’s internal embedding lookup matrix.`,name:"inputs_embeds"},{anchor:"transformers.ConvBertForQuestionAnswering.forward.start_positions",description:`<strong>start_positions</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size,)</code>, <em>optional</em>) — | |
| Labels for position (index) of the start of the labelled span for computing the token classification loss. | |
| Positions are clamped to the length of the sequence (<code>sequence_length</code>). Position outside of the sequence | |
| are not taken into account for computing the loss.`,name:"start_positions"},{anchor:"transformers.ConvBertForQuestionAnswering.forward.end_positions",description:`<strong>end_positions</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size,)</code>, <em>optional</em>) — | |
| Labels for position (index) of the end of the labelled span for computing the token classification loss. | |
| Positions are clamped to the length of the sequence (<code>sequence_length</code>). Position outside of the sequence | |
| are not taken into account for computing the loss.`,name:"end_positions"}],returnDescription:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p>A <a | |
| href="/docs/transformers/pr_49554/en/main_classes/output#transformers.modeling_outputs.QuestionAnsweringModelOutput" | |
| >QuestionAnsweringModelOutput</a> or a tuple of | |
| <code>torch.FloatTensor</code> (if <code>return_dict=False</code> is passed or when <code>config.return_dict=False</code>) comprising various | |
| elements depending on the configuration (<a | |
| href="/docs/transformers/pr_49554/en/model_doc/convbert#transformers.ConvBertConfig" | |
| >ConvBertConfig</a>) and inputs.</p> | |
| `,returnType:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p><a | |
| href="/docs/transformers/pr_49554/en/main_classes/output#transformers.modeling_outputs.QuestionAnsweringModelOutput" | |
| >QuestionAnsweringModelOutput</a> or <code>tuple(torch.FloatTensor)</code></p> | |
| `});var he=e(me,4);h(he,{children:(n,r)=>{var o=u();i(n,o)},$$slots:{default:!0}});var we=e(he,4);m(we,{anchor:"transformers.ConvBertForQuestionAnswering.forward.example",children:(n,r)=>{var o=f(),l=e(c(o),2);p(l,{code:"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",highlighted:`<span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer, ConvBertForQuestionAnswering | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-meta">>>> </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"YituTech/conv-bert-base"</span>) | |
| <span class="hljs-meta">>>> </span>model = ConvBertForQuestionAnswering.from_pretrained(<span class="hljs-string">"YituTech/conv-bert-base"</span>) | |
| <span class="hljs-meta">>>> </span>question, text = <span class="hljs-string">"Who was Jim Henson?"</span>, <span class="hljs-string">"Jim Henson was a nice puppet"</span> | |
| <span class="hljs-meta">>>> </span>inputs = tokenizer(question, text, return_tensors=<span class="hljs-string">"pt"</span>) | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">with</span> torch.no_grad(): | |
| <span class="hljs-meta">... </span> outputs = model(**inputs) | |
| <span class="hljs-meta">>>> </span>answer_start_index = outputs.start_logits.argmax() | |
| <span class="hljs-meta">>>> </span>answer_end_index = outputs.end_logits.argmax() | |
| <span class="hljs-meta">>>> </span>predict_answer_tokens = inputs.input_ids[<span class="hljs-number">0</span>, answer_start_index : answer_end_index + <span class="hljs-number">1</span>] | |
| <span class="hljs-meta">>>> </span>tokenizer.decode(predict_answer_tokens, skip_special_tokens=<span class="hljs-literal">True</span>) | |
| ... | |
| <span class="hljs-meta">>>> </span><span class="hljs-comment"># target is "nice puppet"</span> | |
| <span class="hljs-meta">>>> </span>target_start_index = torch.tensor([<span class="hljs-number">14</span>]) | |
| <span class="hljs-meta">>>> </span>target_end_index = torch.tensor([<span class="hljs-number">15</span>]) | |
| <span class="hljs-meta">>>> </span>outputs = model(**inputs, start_positions=target_start_index, end_positions=target_end_index) | |
| <span class="hljs-meta">>>> </span>loss = outputs.loss | |
| <span class="hljs-meta">>>> </span><span class="hljs-built_in">round</span>(loss.item(), <span class="hljs-number">2</span>) | |
| ...`,lang:"python",wrap:!1}),i(n,o)},$$slots:{default:!0}}),a(pe),a(C);var Ce=e(C,2);ze(Ce,{source:"https://github.com/huggingface/transformers/blob/main/docs/source/en/model_doc/convbert.md"}),J(2),i(ue,B),Ze()}export{Qe as component}; | |
Xet Storage Details
- Size:
- 95.5 kB
- Xet hash:
- f3261912203a4f0fda9039ea20e4bc9c225a60d3a358a02088270c8df83d83b1
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.